Gravar-mail: Fast Approximations of Activation Functions in Deep Neural Networks when using Posit Arithmetic